How it works
How data observability works, end to end
Connect a warehouse, auto-generate monitors, get alerted when data breaks, and track each incident to resolution. Here is exactly what happens at every step.
In one paragraph
Data observability works by continuously reading your warehouse metadata to monitor freshness, volume, schema, and anomalies on every table, building end-to-end lineage from your dbt project, and alerting your team the moment a monitor fires. Each break becomes a tracked incident with root-cause hints and the downstream tables, dashboards, and exports it affects, so you fix the right thing before stakeholders notice.
Connect your warehouse
Authorize read-only access to Snowflake, BigQuery, Databricks, or Redshift. We read metadata and information-schema statistics, not your rows, so the connection is fast and low-risk. Most teams finish in about 5 minutes.
- Read-only metadata access
- Snowflake, BigQuery, Databricks, Redshift
- No rows copied, metadata-first
Auto-generate monitors
We read your dbt manifest and your warehouse schema to auto-create monitors across all 5 pillars on every table, no YAML required. ML-tuned thresholds learn each table's normal behavior so alerts stay high-signal from day one.
- Auto-monitors on every table
- dbt-native, reads your manifest
- ML thresholds learn the baseline
Get alerted when data breaks
When freshness, volume, schema, or distribution breaks, related alerts group into one incident and route to Slack or PagerDuty with severity and ownership. No wall of cron emails, no alert fatigue.
- Slack and PagerDuty
- Grouped, severity-routed alerts
- Low-noise by design
Track to resolution
Each break opens an incident with root-cause hints and the downstream lineage impact. Assign an owner, diagnose with the lineage graph, resolve, and keep the history so you can measure and reduce data downtime over time.
- Auto-opened incidents
- Downstream impact from lineage
- History and MTTR tracking
Architecture
Metadata-first, so monitoring stays cheap and safe
We read warehouse metadata and statistics rather than scanning your tables. That keeps compute cost tiny and means your rows never leave your environment.
Reads metadata, not rows
Freshness, volume, and schema monitors run on information-schema and account-usage statistics, so most checks touch no table data at all.
Sampled, scheduled deep checks
Distribution and null checks sample intelligently and run on a cadence you control, so warehouse credits stay predictable.
Your data stays put
On Enterprise, an in-VPC deployment means data never leaves your environment. Self-serve plans are read-only.
See it live
Watch the whole flow in one console
Alerted #data-eng 0.8s ago.
Downstream impact · consumers at risk
Before you connect
Data observability setup questions
What permissions does a data observability tool need on Snowflake or BigQuery?
A read-only user or service account with access to the schemas you want watched. On Snowflake that is a role with USAGE on the warehouse, database and schemas plus SELECT on the tables; on BigQuery it is the Data Viewer and Job User roles on the project. No write, create or admin grant is needed, and you can scope the user to a single schema to start.
Do I need dbt to set up data observability?
No. Freshness, volume, schema, anomaly and table lineage monitors generate from the warehouse catalog alone, so a team without dbt is monitored the same day. If you do run dbt, upload or sync manifest.json on Team and above and monitors follow your models, sources and owners, which saves the per-table setup on large projects.
What happens when the 14-day trial ends?
Pick a plan and monitoring carries on with the same connection, monitors and history. If you do not pick one, the workspace stops checking tables and shows no data until a plan is active. There is no card on file during the trial, so nothing is charged without you choosing a plan first.
How many tables should a first rollout cover?
Start with the tables behind the dashboards and exports people actually complain about, usually 30 to 100. That fits Starter at up to 50 tables or Team at up to 250, and it gives you real incidents in the first week. Widen coverage once alert routing and owners are settled, so new monitors land on a process that already works.
Explore the platform
Live on day one
Connect a warehouse and watch monitors generate across every table. Transparent pricing, no credit card.